AI pipeline enables dual-site BMD prediction from single KUB radiograph for osteoporosis screening

AI pipeline for osteoporosis screening from a single KUB radiograph showed AUC 0.969, 100% specificity, and 100% PPV in a prospective study (n=51).

• The algorithm, called SHield, predicts bone mineral density (BMD) at both hip and spine from a single kidney-ureter-bladder (KUB) radiograph, integrating dual-site predictions to determine the lowest T-score.
• On internal testing (628 patients), hip and spine models achieved Pearson correlations of 0.887 and 0.921, and AUCs of 0.894 and 0.956 for T-score ≤-2.5.
• The prospective feasibility study used a conservative threshold (Tm-score ≤ -2.8) to minimize false positives, achieving 100% positive predictive value (PPV) and 100% specificity.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.